| """aubin decide case.json | aubin serve --port 8009 | aubin loop "task" [--generator gemma|openai:gpt-4o-mini] [--numeric]
|
|
|
| case.json: {"state": {...}, "questions": {id: {type, instructions, criteria}}}
|
| --think 4 eminsiz sorularda kısa akıl yürütme (0 = kapalı, tek geçiş)
|
| --model "repoA:0.6,repoB:0.4" farklı tabanlardaki üyelerin ansamblı (--ens_T birleşim sıcaklığı)
|
| """
|
| import argparse, json, sys
|
|
|
|
|
| def main():
|
| ap = argparse.ArgumentParser(prog="aubin")
|
| sub = ap.add_subparsers(dest="cmd", required=True)
|
| d = sub.add_parser("decide"); d.add_argument("case"); d.add_argument("--model", default="emrevrg/AUBIN-12B")
|
| s = sub.add_parser("serve"); s.add_argument("--model", default="emrevrg/AUBIN-12B"); s.add_argument("--port", type=int, default=8009)
|
| s.add_argument("--learn", default="", help="öz-öğrenme belleği dosyası (.npz); verilirse /learn etkin, bellek kalıcı")
|
|
|
| lp = sub.add_parser("loop"); lp.add_argument("task"); lp.add_argument("--model", default="emrevrg/AUBIN-12B")
|
| lp.add_argument("--generator", default="gemma"); lp.add_argument("--base_url", default="https://api.openai.com/v1")
|
| lp.add_argument("--n", type=int, default=4); lp.add_argument("--rounds", type=int, default=3)
|
| lp.add_argument("--target", type=float, default=0.8); lp.add_argument("--numeric", action="store_true")
|
| for x in (d, s, lp):
|
| x.add_argument("--perms", type=int, default=1); x.add_argument("--think", type=float, default=None)
|
| x.add_argument("--think_mix", type=float, default=None)
|
| x.add_argument("--ens_T", type=float, default=0.8)
|
| a = ap.parse_args()
|
| from .core import Aubin, AubinEnsemble
|
| specs = [x.rsplit(":", 1) if ":" in x else (x, "1") for x in a.model.split(",")]
|
| ms = [(Aubin(r, perms=a.perms, think_margin=a.think, think_mix=a.think_mix), float(w)) for r, w in specs]
|
| m = ms[0][0] if len(ms) == 1 else AubinEnsemble(ms, a.ens_T)
|
| if a.cmd == "loop":
|
| from .loop import AubinLoop, GemmaGenerator, OpenAIGenerator, last_number, final_line
|
| base = ms[0][0]
|
| gen = GemmaGenerator(base) if a.generator == "gemma" else OpenAIGenerator(a.generator.split(":", 1)[1], a.base_url)
|
| loop = AubinLoop(m, gen, extract=last_number if a.numeric else final_line, target=a.target, max_rounds=a.rounds,
|
| n=a.n, mode="aubin+vote")
|
| print(json.dumps(loop.solve(a.task), ensure_ascii=False, indent=1))
|
| elif a.cmd == "decide":
|
| case = json.load(open(a.case, encoding="utf-8")) if a.case != "-" else json.load(sys.stdin)
|
| print(json.dumps(m.decide(case["state"], case["questions"]), ensure_ascii=False, indent=1))
|
| else:
|
| from .server import serve
|
| if a.learn:
|
| import atexit, os
|
| from .learn import AubinLearning
|
| m = AubinLearning(m)
|
| if os.path.exists(a.learn):
|
| m.load(a.learn)
|
| atexit.register(lambda: m.save(a.learn))
|
| serve(m, a.port)
|
|
|
|
|
| if __name__ == "__main__":
|
| main()
|
|
|
